app.py

import os
import torch
from transformers import LayoutLMv3ForTokenClassification, LayoutLMv3FeatureExtractor, LayoutLMv3Tokenizer
from PIL import Image
import pytesseract
from pdf2image import convert_from_path
import re

# Ensure you have the necessary dependencies installed:
# pip install transformers torch Pillow pytesseract pdf2image

# Set up pytesseract path (adjust as needed)
pytesseract.pytesseract.tesseract_cmd = r'/usr/bin/tesseract'

# Load pre-trained model and tokenizer
model_name = "microsoft/layoutlmv3-base"
model = LayoutLMv3ForTokenClassification.from_pretrained(model_name, num_labels=5)  # Adjust num_labels as needed
feature_extractor = LayoutLMv3FeatureExtractor.from_pretrained(model_name)
tokenizer = LayoutLMv3Tokenizer.from_pretrained(model_name)

# Define label mapping
id2label = {0: "O", 1: "COMPANY", 2: "EDUCATION", 3: "POSITION", 4: "DATE"}
label2id = {v: k for k, v in id2label.items()}

def preprocess_document(file_path):
    if file_path.lower().endswith('.pdf'):
        images = convert_from_path(file_path)
        image = images[0]  # Process only the first page for simplicity
    else:
        image = Image.open(file_path)
    
    # Perform OCR
    ocr_result = pytesseract.image_to_data(image, output_type=pytesseract.Output.DICT)
    
    words = ocr_result['text']
    boxes = []
    for i in range(len(words)):
        x, y, w, h = ocr_result['left'][i], ocr_result['top'][i], ocr_result['width'][i], ocr_result['height'][i]
        boxes.append([x, y, x+w, y+h])
    
    return image, words, boxes

def process_resume(file_path):
    image, words, boxes = preprocess_document(file_path)
    
    # Prepare inputs for the model
    encoding = feature_extractor(image, words, boxes=boxes, return_tensors="pt")
    input_ids = encoding["input_ids"]
    attention_mask = encoding["attention_mask"]
    token_type_ids = encoding["token_type_ids"]
    bbox = encoding["bbox"]
    
    # Forward pass
    outputs = model(input_ids=input_ids, attention_mask=attention_mask, token_type_ids=token_type_ids, bbox=bbox)
    
    # Get predictions
    predictions = outputs.logits.argmax(-1).squeeze().tolist()
    
    # Post-process results
    parsed_info = {"COMPANY": [], "EDUCATION": [], "POSITION": [], "DATE": []}
    current_entity = None
    current_text = ""
    
    for word, label_id in zip(words, predictions):
        if label_id != 0:  # Not 'O' label
            label = id2label[label_id]
            if label != current_entity:
                if current_entity:
                    parsed_info[current_entity].append(current_text.strip())
                current_entity = label
                current_text = word + " "
            else:
                current_text += word + " "
        else:
            if current_entity:
                parsed_info[current_entity].append(current_text.strip())
                current_entity = None
                current_text = ""
    
    return parsed_info

def main():
    resume_path = input("Enter the path to your resume file (PDF or image): ")
    if not os.path.exists(resume_path):
        print("File not found. Please check the path and try again.")
        return
    
    parsed_info = process_resume(resume_path)
    
    print("\nExtracted Information:")
    print("Companies worked for:", ", ".join(parsed_info["COMPANY"]))
    print("Education:", ", ".join(parsed_info["EDUCATION"]))
    print("Positions held:", ", ".join(parsed_info["POSITION"]))
    print("Relevant dates:", ", ".join(parsed_info["DATE"]))

if __name__ == "__main__":
    main()